Getting very poor performance on MNIST database

I’m using resnet34 to train my fast.ai model to train it on the MNIST database, but I’m getting very poor results. Any idea how to use other architectures like Lenet-5 on fast.ai?